Type | |
Stats | 114 461 |
Reviews | (37) |
Published | Nov 16, 2024 |
Base Model | |
Training | Steps: 432 Epochs: 10 |
Usage Tips | Clip Skip: 1 |
Trigger Words | haxorus |
Hash | AutoV2 14DC23EEED |
Haxorus LoRA trained on Pony upon request. Another low image count LoRA that punches way above its weight class. Very happy with how this one turned out. It has some issues to work around, which I'll list below, but it's overall quite pleasant to work with.
Issues:
Scale color patterns can be inconsistent. This is due to the varying depictions of Haxorus in the source data. The yellow and grey spots on the body get distributed differently
Add some reinforcement tags to your prompt to help address this. I put (grey torso, grey breasts) in my positive and (yellow breasts, yellow torso) in my negative. This improved consistency a ton
Scale patterns can gen on top of tight fitting clothing
If you're prompting for scale appearance like I was, remove that from your prompt when genning with clothing
Generation Tips:
Use 'scalie' instead of 'furry' in your prompt for more consistent results